A Structure Based Drug Design pipeline involves the development of drug molecules or ligands that form strong complexes with a given receptor at it's binding site. A prerequisite to this is the requirement of finding druggable binding sites on the 3D structure of the protein. Although several methods for detecting binding sites have been developed beforehand, they sometimes break down due to failures in identifying binding cavities and ranking identified cavities accurately. The rapid adoption and success of Deep Learning algorithms in various sections of structural biology beckons the usage of such algorithms for binding site detection. In this work, we design a novel framework called DeepPocket that utilises 3D Convolutional Neural Networks for the scoring and segmentation of identified cavities on the protein surface. We show that this framework outperforms all the current state-of-the-art methods in ligand binding site identification and segmentation.
Rishal Aggarwal
Akash Gupta
Jawahar C.V.
Deva Priyakumar U.
Deep Learning, Drug Discovery, Structural Biolog.
Algorithm.
Technology designed and implemented.
Drug Design, Healthcare.